I'm sorry for resurrecting an old topic, but I figured it's worth posting since I couldn't find any similar solution on the forum. I hope someone finds this useful while S7-1500 with native double float is gaining traction.
If I understood e3's post correctly, he saw similar problem in R2_ADD2 function as we did. My colleague "mme" came up with this solution based on R2_ADD:
tR2_ADD2.RX = Y.R1 + Y.RX + X.RX + X.R1;
tR2_ADD2.R1 = X.RX + Y.RX - tR2_ADD2.RX + Y.R1 + X.R1
I did a quick and dirty test in C# using floats in place of S7 Real and with double as a reference. The code above appears to be working as expected.
Test code:
using System;
using System.Collections.Generic;
using System.Linq;
using System.Text;
using System.Threading.Tasks;
namespace Real2Test
{
class Program
{
static void Main(string[] args)
{
float a = 10000000F;
float b = 0.00123F;
double c = a;
Real2 d = new Real2();
d.RX = a;
d.R1 = 0F;
Real2 e = new Real2();
e.RX = b;
e.R1 = 0F;
for (int i = 0; i < 3333333; i++)
{
c -= (double)b;
d = R2AddNew(d, R2Mul(e, -1F));
}
Console.WriteLine("Double: {0:F8}\tR2: {1:F8}\t{2:F8}", c, d.RX, d.R1);
for (int i = 0; i < 3333333; i++)
{
c += (double)b;
d = R2AddNew(d, R2Mul(e, 1F));
}
Console.WriteLine("Double: {0:F8}\tR2: {1:F8}\t{2:F8}", c, d.RX, d.R1);
for (int i = 0; i < 3333333; i++)
{
c += (double)b;
d = R2AddNew(d, R2Mul(e, 1F));
}
Console.WriteLine("Double: {0:F8}\tR2: {1:F8}\t{2:F8}", c, d.RX, d.R1);
Console.ReadKey();
}
static Real2 R2Mul(Real2 x, float y)
{
Real2 tmp = new Real2();
tmp.RX = x.RX * y;
tmp.R1 = x.R1 * y;
return tmp;
}
static Real2 R2AddNew(Real2 x, Real2 y)
{
Real2 z = new Real2();
z.RX = y.R1 + y.RX + x.RX + x.R1;
z.R1 = x.RX + y.RX - z.RX + y.R1 + x.R1;
return z;
}
}
class Real2
{
public float R1 { get; set; }
public float RX { get; set; }
}
}
Thank you for this great library!
W
Thema: Arithmetik doppelter Genauigkeit (Gelesen 19463 mal)